LABOUR MARKET HETEROGENEITY: DISTINGUISHING BETWEEN UNEMPLOYMENT AND INACTIVITY*

Transcription

1 LABOUR MARKET HETEROGENEITY: DISTINGUISHING BETWEEN UNEMPLOYMENT AND INACTIVITY* Mário Centeno** Pedro Afonso Fernandes*** 1. INTRODUCTION In economic analysis, the unemployment rate and the participation rate are two of the most frequently used indicators. Defining the concept of unemployment is therefore an important issue, both for research in economics and for the definition of public policies. Most countries distinguish between the unemployed and the inactive by using a criterion relating to employment search. This definition, however, does not make it possible to highlight the differences existing within each group, above all among those classified as out of the labour force. The purpose of this article is to describe this heterogeneity and its implications in the dynamics of the active population. An attempt will be made, therefore, to give an answer to the following question: is it possible to classify the non-employed in a way that makes it possible to create homogeneous groups in terms of their degree of labour market attachment? Seen from the angle of how the labour market works, it is important to distinguish between those who are looking for work (the unemployed) and those not looking (the inactive). This is particularly important for analyses based on the flows between the different labour market states (normally between employment, unemployment and inactivity). From this point of view, the idea of looking for a job is very often replaced by the notion of * The views expressed in this article are those of the authors and not necessarily those of the Banco de Portugal. ** Economic Research Department. *** Centro Interdisciplinar de Estudos Económicos. productive waiting for new work (see, for example Blanchard and Diamond, 1992). The relevant distinction ceases to be based on the actions that the unemployed undertake in their search for work, but rather on their productivity during non-employment periods and assessed by the rate of transition to work. This view of the labour market and the way it works is sustained by the importance of transition between inactivity and employment. In Portugal, for example, this represents between 1 and 2 percent of the active population for each three-month period. Given this, we could well think that the behaviour of some individuals who are not looking for work but who are prepared to work, should be considered close enough to the concept of activity such that they could even be included in the group of the unemployed (1). If the concept of unemployed is based on the search-for-work criterion, we are assuming that the effort to find work reveals a close connection between individuals and the labour market (2).On the other hand, those who are not looking for work, even though they might want to work, are (1) The recent debate on the inclusion of the discouraged (those who are not looking for work because they think it is difficult to find work) in the concept of the unemployed (OECD, 1987 and 1995) is an example of this approach. (2) The definition of the International Labour Organisation and Eurostat distinguishes between active and passive attitudes in the search for work. An example of the first is contact with employers and an example of the second is looking through the job ads. Banco de Portugal / Economic bulletin / March

2 considered not to be showing an attachment that is sufficiently strong with the activity status to justify their classification as unemployed. The present study uses this definition of the active population as a starting point to look at the homogeneity of the group of individuals classified as inactive, namely to analyse a group that is marginally attached, consisting of those who would like to work but who are not looking for work (3). In most countries this group is small in terms of the whole active population, but usually it represents more than 30 per cent of the unemployed. In the work of Jones and Riddle (1999) and (2000), the marginally attached account for between 25 per cent and 30 per cent of the unemployed in Canada, but they reach two-thirds in the USA. Portugal stands midway between these figures. Between 1992 and 2003, the marginally attached represented, on average, 30 per cent of male unemployed and 50 per cent of female. However, it is the dynamics of the transition to work that is the most important factor in the context of this article. Our aim therefore is to characterise the level of homogeneity in the different states of the labour market through a comparison of the transition rates for individuals belonging to each state. This analysis provides a contribution towards the definition of unemployment but it also leads to a better understanding of the cyclical behaviour in particular of Portuguese labour market aggregates. Above all, it helps us to understand the procyclical behaviour of the participation rate and its contribution to the persistence of unemployment through the cycle. The transitions to inactivity imply that in a recession the average unemployment rate increases more slowly than expected compared with the way employment is moving. In the same way, during periods of expansion, unemployment tends not to fall as quickly as expected, given the behaviour of those who are inactive and become newly unemployed during these periods. Methodologically, this work is a development of the approach used by Jones and Riddel (2000) and (2002), in terms of attachment to the labour (3) In the Inquérito ao Emprego (IE), the Portuguese labour market survey, as in most of the surveys in OECD countries, readiness to work and in search of work are collated in relation to the period when the survey is being undertaken. market. Their work revealed the wide heterogeneity that characterises those who are not in work in Canada and the United States. What this means is that there are wide differences between the unemployed, the inactive and some sub-categories of both when it comes to the actual situation of future transitions between states of the labour market, specifically in the movements towards finding employment. The analytical model is tested for Portugal using data of the Inquérito ao Emprego (IE) for the period between the second quarter of 1992 and the fourth quarter of This approach reveals, among other things, the differences in the way the inactive behave in terms of attachment to the labour market. Above all, the marginally attached behave very differently from the rest of the inactive group and seem in many cases equivalent to the unemployed. These results justify the separation of the state of marginally attachment, given the higher probability of finding work in the future of individuals in this state. Indeed, in the period under review, the rate of transition to employment in this group is similar to the unemployed and clearly higher than the remainder of the inactive group. Given this evidence, an argument could be put forward for using three aggregates for those out of work: the unemployed, the marginally attached and the inactive. One advantage of this approach would be to make it easier to identify the level of competition between those who are not employed when it comes to filling job vacancies. This is not satisfactorily reflected in the unemployment rate. The transition rates of the marginally attached will be enough to justify the above classification, as a group of non-employed who compete productively for the employment offers (necessarily limited) available in the economy. The lower the rate of job offers in the economy the grater will be the importance of these results for an understanding of the matching process in the labour market. In fact, the relevance of the criterion of active search for a job should be greater in economies with lower rates of job offers (as is the case in Portugal). In a situation where there is job rationing, the differences in the transition rates between unemployed and inactive individuals are expected to be clearer. However, this behaviour may not surface if there are negative self-selection 62 Banco de Portugal / Economic bulletin / March 2004

3 processes for the state of unemployment. This self-selection may be associated with a deterioration in the overall unemployment situation, which becomes visible either in terms of employability (where the active search for a job is justified by systematic difficulties in finding work) or through inefficiency in the search for work. The latter argument has been extensively used by public authorities to justify the implementation of a wide range of search support programmes in European Union countries. 2. DATA The source of information for our empirical work are the individual records (duly rendered anonymous) in the IE relating to the period between the second quarter of 1992 and the fourth quarter of This covered individuals between the ages of 16 and 64. The survey is carried out every quarter, with a rotation scheme in the procedure (one-sixth for each quarter). It respects conceptual norms, methodology and quality common to most countries in Europe. The sample for each quarter is made up of around 40,000 individuals. Generally speaking, there are six records for each individual. The fact that information is available over a long period means that for most individuals it is possible to observe the transition between states for five consecutive quarters. This short panel characteristic of the survey means that it is possible to estimate duration models for unemployment and inactivity that will be used to test equality in the rates of transition between the unemployed, the marginally attached and the other inactive individuals. It is important to make a remark on the data that is used. The estimation of duration models requires the computation of an approximate figure for the duration of inactivity, and this is not supplied directly through the survey. In the case of left-censured observations (4), the duration of inactivity is deemed to be identical to the duration of the period of non-employment since the end of the previous employment spell, if applicable, or to the estimated period of time since the individual left (4) Records relating to individuals who are inactive the first time they are observed. school (and this depends on the individual schooling), in the case of individuals who have never worked. In those observations, which are not left-censured, the duration of inactivity is computed using the information over the 6 successive quarters in which each individual remains on file. 3. METHODOLOGY The methodology used takes advantage of the longitudinal information, and tests whether two (or more) states of non-employment are identical in behavioural terms. It was originally formulated by Flinn and Heckman (1982) and (1983). The idea underlying the Flinn and Heckman test is very simple: with accurate control of the characteristics of the individual, if the transition rate from state x to state z is identical to the transition rate from state y to state z (for all the state of destiny z) then the state of origin (x or y) is irrelevant in terms of determining the transition rate of individuals for (all of) z. Using a small sample of qualified young North American males (taken from the National Longitudinal Survey of Young Men), Flinn and Heckman rejected the hypothesis that the behaviour of the unemployed and the inactive was identical. Similar results, also for the United States, were obtained by Tano (1991) and Gönül (1992) for wider groups (young people and adults of both genders in one case and young people of both genders in the other). This application uses a model with four states: employment (E), unemployment (U), and inactivity is divided into two states: marginal attached (M) and other inactivity (N). The marginally attached are distinguished from the other inactive because of their availability to work and from the unemployed because they do not search for work. In this model, the states of unemployment and marginal attachment will be equivalent if: The transition rate from U to E is the same as the transition rate from M to E. h UE h ME The transition rate from U to N is the same as the transition rate from M to N. Banco de Portugal / Economic bulletin / March

4 h UN h These equalities test whether unemployment (U) and marginal attachment (M) can be considered identical behavioural states (Hypothesis A). In this case, the activity of job-seeking carried out by the unemployed but not by the marginally attached is not more productive than the availability to work (common to both states). In other words, the distinction between unemployment and non-participation can be made solely on the basis of the availability to work. The distinction between marginal attachment and inactivity will be irrelevant if: MN The transition rate from M to E is the same as the transition rate from N to E. h ME h NE The transition rate from MtoUis the same as the transition rate from N to U. h MU h These equalities test whether the marginal attachment (M) and other inactivity (N) can be deemed to be identical behavioural states (Hypothesis B). This situation corresponds to the usual classification of individuals in terms of employed, unemployed and inactive (E, U or O where O=M+N). Here the availability for work does not incorporate any additional information vis-à-vis attachment to the labour market. Of course, if all the distinctions make sense, a model with four states (E, U, M and N) would be a better description of the labour market than the usual one that uses three states (E, U, and O). Even when these conditions of equivalency between states are put aside, it could still make sense to separate the states in terms of their attachment to activity. This is what happens if, for example, h h h UE ME NE ; huu hmu hnu and hun hmn hnu. In this case, M is an intermediate state, with a level of attachment which is greater than that of the other inactive (N) although less than the unemployed (U). NU 4. ECONOMETRIC MODEL The two hypotheses outlined above were tested using two kinds of analysis. Firstly, we compute empirical transition rates between the various states of the labour market; and secondly, we estimate an econometric model in which we condition these transitions on observed individual characteristics and the length of the experiences among the non-employed. In the second model, a likelihood ratio test was applied. This compares the maximum values of the logarithm of the likelihood function under the null hypothesis of states equivalence LH o, and under the alternative hypothesis of the states being differentiated LH a. This test is based on the following statistic (Griffiths et al. 1993): 2 LH LH LR 2 a o ~ (1) R where R is the number of restrictions under the null hypothesis. For the two hypotheses we are looking at (A and B), we estimate (by gender and year) two variants of a duration model: the first is associated to the null hypothesis (the restricted model) where the regression coefficients, referring to the variables among individuals belonging to the two states of origin in the test (5), were forced to equality; and a second variant, relating to the alternative hypothesis (the non-restricted model) in which the regression coefficients may vary freely between the two origins. Then the respective logarithm values of the likelihood function were used to calculate statistic (1), checking the behavioural identity of the two origins through the comparison of this statistic with the critical values of the 2 distribution. The number of degrees of freedom was identical to the number of coefficients, which, in the restricted model, did not vary freely vis-à-vis the non-restricted model. 5. EMPIRICAL RESULTS Table 1 shows the average transition rates in periods of economic recession and expansion; Charts 1 and 2 show the exit rates into employ- (5) U and M in Hypothesis A; M and N in Hypothesis B. 64 Banco de Portugal / Economic bulletin / March 2004

5 Table 1 AVERAGE TRANSITION RATES IN PERIODS OF RECESSION AND EXPANSION Transition to employment Period Males h UE h ME h NE Transition to unemployment h UU h MU h NU Transition to other inactivity h UN h MN h NN Transition to employment Females h UE h ME h NE Transition to unemployment h UU h MU h NU Transition to other inactivity h UN h MN h NN Source: Inquérito ao Emprego (IE). ment among the various sub-categories of the population, for males and females respectively (6),(7). Some features of these results are worth a special comment. Firstly, in spite of the fact that the transition rates are relatively stable over time, there is a certain tendency for increase in the transitions between M and E. The stability of the transitions does not completely eliminate their cyclical behaviour, shown, for example, in the fall in transitions to employment in 1995 and These are the two years in the period covered by the survey when the output gap reaches its minimum values in each of the two economic recession phases. In addition, the transition rates order remains the same for each of the periods considered hue hme hne ; huu hmu hnu and hun hmn hnn. It should be noted that the difference between h ME and h NE is greater than the difference between h UE and h ME for both genders and in every year analysed. These results were obtained without conditioning the transitions in the characteristics of the individuals and they suggest that the state of marginal attached is closer to the state of unemployed than to the state of other inactive. Finally, the difference between the rates of transition for employment (from M and U) is subject to some cyclical fluctuation (see Charts 1 and 2). Generally speaking, the difference is greater in periods of recession than in expansion. This must stem from both the composition of the two groups (which is reflected in the cyclical behaviour of unemployment) and also from the normal cut back in the number of job vacancies in a recession. This in its turn makes the criterion of looking for work more relevant in determining transition rates. Overall, these results indicate that category O (M +N) is very heterogeneous. This goes above all for those not looking for work in the period covered by the survey, but who are available to work. Their behaviour is clearly different from those who do not want to work and are not looking for employment. We then studied to what point the results of the unconditional analysis are affected by consider- (6) The employment survey questionnaire changed in 1998 and this led to a significant change in the intensity of some of the estimated flows (particularly among the non-employed states). The graphs indicate this discontinuity with a separation in the curves for the point. The analysis in Table 1 takes the change in the economic cycle from the farthest points of the recession (1995) to the top of the boom (2000). Given the break in the survey s questionnaire, the boom period is divided at (7) The transition rates in Table 1 were calculated as follows: d ij hij ri where d ij Number of individuals in state i in the initial period who transfer to state j in the subsequent period; ij, EU,, MN, (with the normal interpretation of the notation). Banco de Portugal / Economic bulletin / March

6 Chart 1 QUARTERLY TRANSITION RATES INTO EMPLOYMENT Male Chart 2 QUARTERLY TRANSITION RATES INTO EMPLOYMENT Female hue hue In percentage of the original state hme In percentage of the original state hme hne hne Source: Inquérito ao Emprego (IE). Source: Inquérito ao Emprego (IE). (8) The variables used in the restricted model relate to the personal characteristics of the individuals: age, marital status (1 for yes, 0 for no), head of household (1/0), secondary education (as a maximum level of qualifications: 1/0), higher education (idem), of foreign nationality (1/0), in receipt of benefit (subsidy) for unemployment (1/0), duration of last job, end of contract as the reason for losing the last job (1/0), sacked or suspended from the last job (1/0) and the region of residence. (9) The p-value is a statistic from the test undertaken and it gives us an indication of the probability of the null hypothesis being rejected ation of the observable characteristics of the individuals. The previous analysis allows us to answer the question as to whether the existing measurement of unemployment is appropriate. But to answer the more general question as to the appropriateness of this measurement for any group of individuals, it is important to set out the conditions for the results of observable characteristics among the sample used. The results of the equivalence tests on states U, M and N are presented in Table 2 (8). This Table shows the p-values (9) associated with each of the likelihood ratio tests of equation (1). By and large, these results confirm what a simple view of Charts 1 and 2 might suggest. The equivalence between the states of marginally attached and other inactivity is completely rejected, both for males and for females. In every year of the sample under review, the availability for work is important to measure the degree of attachment with the active population and conditions the observed transitions in a significant way. From this point of view, inactivity can be seen as a heterogeneous state. The tests undertaken with the estimated duration model for the equivalence between U and M lead to a clear rejection of the null hypothesis in most of the years for females. For males the null hypothesis is accepted in 1995, 1997, 1998 and 2003 (with a level of significance of 10 per cent). The results for males point to a proximity of the U and M states that is not evident in the female case, in which the two states seem to have more clearly separate dynamics. Overall, these results reveal that being marginally attached is a distinctive state in the labour market. It would therefore be desirable its measurement, publication and analysis to be carried out on a more regular basis. In this respect, the study of flows between M and the other states would be particularly important. These results are similar to those obtained in other countries that can serve as a comparison, specifically the USA and Canada. The studies of Jones and Riddel show a greater attachment between unemployment and marginal inactivity in Canada than in the USA, where those who are marginally inactive are distinct from the other inactive. It is not possible, however, to accept the hypothesis of equivalence with the unemployed. In 66 Banco de Portugal / Economic bulletin / March 2004

7 Table 2 RATIO TESTS FOR HYPOTHESES A (U=M?)EB(M=N?) ( ) Gender and year Number of observations Hypotheses A (U=M?) 2 Degrees of freedom P-value Number of observations Hypotheses B (M=N?) 2 Degrees of freedom P-value Males *** ** ** *** *** *** *** *** *** *** * *** ** *** *** *** ** *** ** *** *** Females *** *** *** *** *** *** *** *** *** *** *** *** *** *** *** ** *** *** *** *** *** * *** Source: Inquérito ao Emprego (IE). Note: ***, ** and * denote de rejection of the null hypothesis at the 1%, 5% and 10%, levels of significance, respectively. Canada, this hypothesis can be accepted for some years, but less frequently than for Portugal. 6. FINAL REMARKS The aim of this work is to question whether it is valid to separate the unemployed from the inactive for statistical purposes. The empirical approach of the present analysis is based on the detailed study of how certain subcategories of the inactive behave, with special reference to the marginally attached, that is, those out of the labour force who are not looking for employment but are prepared to work and available if it should come. Above all, the attempt was made to test whether the state of marginally attached (M) is identical to the state of unemployed (U), from the behavioural point of view and to the other form of inactivity (not marginal: N). To do this, reference was made to the longitudinal data of the Employment Survey covering 1992 to This took in more than one complete economic cycle. One of the main results of the analysis is the rejection of the hypothesis of equivalence between the marginally attached and other inactive individuals. The statistical tests lead to the unequivocal rejection of equality between M and N (as between U and N) in all of the twelve years studied. The concept of non-participation therefore has behavioural significance, but there is a wide heterogeneity in the group that is normally identified with inactivity. The availability to work repre- Banco de Portugal / Economic bulletin / March

8 sents more than a simple wish: it implies the existence of some attachment to the labour market and means a significant increase in the likelihood of transferring into employment in the future. The second result of this analysis is that the category of marginally attached is much closer to the state of unemployed than to the other inactive. Mainly for men, but also for women, the hypothesis of equivalence between U and M cannot be rejected for several years of the analysis. This result points to consideration of the group available for work as an intermediate group, whose behaviour is mid way between the unemployed and other inactive. From this point of view, it would be an advantage to have statistics that reflect the four aggregates of the population and to have a separation between the three groups of non-employed in analyses of labour market flows. These results are along the lines of the existing evidence from other countries, especially the US and Canada. In these countries, however, the states U and M do not show the degree of attachment seen in the Portuguese labour market. REFERENCES Blanchard, O.J. and P. Diamnod (1992), The Flow and Approach to Labor Markets, American Economic Review (Papers & Proceedings), Vol. 82, no. 2, May, pp Flinn, C.J. and J.J. Heckman (1982), New Methods for Analysing Structural Models of Labor Force Dynamics, Journal of Econometrics, no. 18, pp Flinn, C.J. and J.J. Heckman (1983), Are Unemployment and Out of the Labor Market Behaviorally Distinct Labor Force States?, Journal of Labor Economics, Vol. 1, no. 1. Gönül, F. (1992), New Evidence on Whether Unemployment and Out of the Labor Force are Distinct States, Journal of Human Resources, no. 27, pp Griffiths, W.E., R.C. Hill and G.G. Judge (1993), Learning and Practicing Econometrics, John Wiley & Sons, Nova Iorque. Jones, S.R.G. and W.C. Riddell (1999), The Measurement of Unemployment: An Empirical Approach, Econometrica, Vol. 67, no. 1, January, pp Jones, S.R.G. and W.C. Riddell (2000), The Dynamics of US Labor Force Attachment, Contributed Paper 0011, Econometric Society World Congress Jones, S.R.G. and W.C. Riddell (2002), Unemployment and Non-Employment: Heterogeneities in Labour Market States, Department of Economics Working Paper Series, , McMaster University, Hamilton, Ontario, Canadá. Kalbfleisch, J.D. and R.L. Prentice (1980), The Statistical Analysis of Failure Time Data, John Wiley & Sons. OECD (1987), On the Margin of the Labour Force: An analysis of discouraged workers and other non-participants, Employment Outlook, September, OECD (1995), Supplementary Measures of Labour Market Slack, Employment Outlook, July, Tano, K.T. (1991), Are unemployment and out of the labor force behaviorally distinct labor force states?, Economic Letters, no. 36, pp , North-Holland. 68 Banco de Portugal / Economic bulletin / March 2004